AMD Ryzen AOCC 1.0 Compiler Tuning Benchmarks
With FFTW you can see the impact of no optimizations (-O0) versus the common -O2 level generally used by most Linux distributions. In the case of FFTW, the strided-vectorization option only had a small impact on performance as did using -Ofast.
SciMark2 saw small benefit out of -O3 optimizations while the additional tuning did little to help this C micro-benchmark.
The znver1 tuning was of little help to the Monte Carlo sub-test.
TSCP doesn't end up being much faster beyond -O2.
The same goes for GraphicsMagick although the potentially-unsafe -Ofast optimizations did boost the performance a bit.